What this category of data looks like
Data for coal mining production policy and regulatory reference comes from official normative documents released by the National Energy Administration, Ministry of Emergency Management, and the Coal Industry Standardization Technical Committee, as well as internal safety production rules and job operation procedures from individual coal mining enterprises. Update frequency: National policies are revised every quarter to annually, while internal rules are updated as needed. Document structure includes fields such as release document number, issuing authority, effective date, clause hierarchy (article, paragraph, sub-paragraph), applicable scenario, and penalty clauses for violations. Most units use national standard measurement units, and policy document numbers follow standard formats like GB, AQ.
Constraints Imposed on Knowledge Base Retrieval and Recall
Official policy documents have rigorous, legally binding clauses. Retrieval must avoid matching irrelevant or inaccurate content, so a high matching threshold is required. Single policy documents can contain dozens of clauses. When chunking content, retain full clause integrity, and do not split content across articles. Policies have clear effective dates and applicable scopes. Add time filter conditions during retrieval to ensure only currently valid policy content is returned. External policies and internal rules coexist. Isolate them via dataset classification to prevent interference between different data categories. Policy terms are professional and fixed. Do not overly rewrite original queries, as this may deviate from official standard expressions.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Coal mining policy clauses are rigorous, require high matching accuracy to avoid referencing irrelevant clauses by mistake |
limit | Top 8 results | Single policy clause has long content, too many results will exceed context window limits |
usingReRank | Enabled | Policy document content is professional, re-ranking is needed to filter the most relevant clauses for the query |
datasetSearchUsingExtensionQuery | Disabled | Policy terms are fixed, rewriting queries may deviate from official standard expressions |
searchMode | Exact match | Policy clauses must strictly match document numbers or clause IDs, avoid errors from fuzzy matching |
collectionIds | Bind exclusive dataset ID for coal mining policies | Differentiate external policies and internal rules, isolate retrieval data across categories |
Three Common Misconfigurations
- Symptom: A re-ranking model is configured, but re-ranking is marked false in Q&A response results. Cause: The
usingReRankswitch was not enabled in the retrieval configuration for the target dataset, or no valid re-ranking model was bound. - Symptom: No optional values appear when referencing knowledge base variables in a workflow. Cause: Dataset access permissions were not enabled in the global configuration, or the
collectionIdsparameter was not specified to bind the target dataset in the configuration. - Symptom: Retrieved policy clauses contain content split across multiple articles. Cause: Clause integrity was not preserved during chunking, with different parts of the same clause split into multiple chunks.
How to Verify Proper Configuration
- Access the dataset management page of FastGPT v4.17.0, select the exclusive coal mining policy dataset, check the
usingReRankswitch status in the retrieval configuration, and confirm it is enabled. - Initiate a test retrieval, input a specific policy-related question, review the matching degree of the retrieval results, and confirm the
similaritythreshold meets expectations. - Check the workflow configuration, confirm the
collectionIdsparameter is specified and the correct coal mining policy dataset ID is bound, and corresponding optional values appear when referencing variables. - Review the retrieval logs, confirm the returned results include the re-ranked sorting field and are marked true, which aligns with the configured requirements.
The configuration values listed above are common starting points. Adjust them based on your own specific samples and use cases.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.